基于高光谱的柑橘叶片磷含量估算模型实验

以117株园栽罗岗橙为实验对象,分别在壮果促梢期和采果期2个不同发育阶段采集234个数据样本,高光谱反射数据构成每个数据样本中的高维矢量描述,用化学方法测得磷含量值作为样本真实目标值,用偏最小二乘法(PLS)及支持矢量回归(SVR)2种多元回归分析算法,在对反射光谱进行各种形式预处理的基础上对柑橘叶片磷含量进行建模和磷含量预测.模型分别在校正集和测试集上进行评估,取得最佳模型决定系数分别为0.905和0.881,均方误差分别为0.005和0.004,平均相对误差分别为0.026 4和0.031 2.实验结果表明:基于高光谱反射数据进行磷含量预测是可行的.

Hyperspectral Estimation Model of Total Phosphorus Content for Citrus Leaves

Field experiments were conducted on 117 planted Luogang citrus trees in the crab village of Guangzhou. 234 pairs of data sample were collected in two different development stages, respectively, germination period and fruit picking period. Hyperspectral reflection data was used as high dimensional vector description. Phosphorus content measured by chemical method as true label and to predict the phosphorus content of citrus leaves. Two mainstream multivariate regression analysis algorithms, partial least squares and support vector regression, were used for modeling and prediction after various preprocessing on spectral reflectance data. Calibration and validation sets were used to evaluate the predictive performance of model. Two regression analysis methods respectively achieved coefficient of determination of 0.905 and 0.881, the MSE of 0.005 and 0.004, the mean relative error of 0.0264 and 0.0312, respectively. The experimental results showed that it is an effective way to predict phosphorus level based on hyperspectral reflection data.